Research On Intelligent Collision Avoidance Decision-making Technology For Autonomous Ship At Sea

With the continuous development and application of intelligent ship technology,in recent years,maritime autonomous ships have gradually become a research hotspot in the maritime field both at home and abroad.One of the important features of maritime autonomous ships is“ships that can operate autonomously without human intervention”,which puts forward higher requirements for the collision avoidance decision-making technology.Therefore,this dissertation conducts a series of in-depth research on the core technical points of intelligent collision avoidance decision-making for autonomous ships at sea,as follows:(1)In this dissertation,the calculation model of ship collision risk based on domain violation parameters is proposed.The model is based on a set of spatio-temporal parameters:the degree of domain violation and the time to domain violation.By introducing the ship position prediction uncertainty model and integrating the ship’s route information in the parameter calculation process,the accuracy of collision risk calculation is improved,and it can further meet the requirements of maritime autonomous ships for the identification accuracy of collision risk between ships.This model optimizes the calculation method of domain violation parameters.First,by analyzing the nature of ship’s position offset,it can be decomposed into the cumulative error value of the horizontal and vertical velocity vector along the ellipse field with time.Based on the Wiener process,the uncertainty of ship position prediction is modeled,the confidence ellipse of ship position uncertainty is constructed using chi-square distribution,and a new ellipse field is constructed by fusion;after that,by analyzing the space-time relationship of mutual domain violation,four types domain violation scenarios are proposed,and the parameter calculation methods under different scenarios are proposed;finally,in view of the fact that the ship’s navigation parameters will change near the turning point,the ship’s route information is integrated into the parameters for the first time using the idea of segmented solution during the calculation process,the accuracy of the domain invasion parameter calculation results has been greatly improved.In addition,this dissertation also draws on the idea of “field theory”,abstracts the degree of domain violation as “the charge quantity of the field source center”,and abstracts the time to domain violation as “the distance from the field source center”.For the expression of space-time correlation characteristics between domain violation parameters,a collision risk calculation model is constructed.The rationality and accuracy of the model are verified through the analysis of multiple groups of comparative experiments.(2)From the perspective of “own ship”,this dissertation proposes the collision avoidance decision-making method under the constraints of multiple navigation feature elements.Based on the thinking of the real ship collision avoidance scenarios at sea,by introducing the constraints of various navigation characteristic elements during the ship collision avoidance decision-making process,the collision avoidance decision-making can better meet the requirements of the actual collision avoidance process,and further improve the completeness of the collision avoidance decision-making algorithm.First of all,this method constructs the initial collision avoidance decision space based on the improved velocity obstacle method,including the core velocity obstacle space related to the dangerous encounter distance and the broad velocity obstacle space related to the safe encounter distance;Hereafter,the constraints caused by various navigation characteristic elements on the decision space for collision avoidance are analyzed,including the maneuverability constraints caused by the nonlinear motion characteristics of ship,the multi-ship constraints caused by complex encounter situations at sea,the rule constraints generated by the “International Regulations for Preventing Collisions at Sea”,and the off-course constraints generated by reducing the deviation in order to improve the collision avoidance efficiency,finally,the optimization of the collision avoidance decision space is realized.Afterwards,based on the optimized decision space and combined with the finite state machine modeling theory,a dynamic behavior response model of the target ship in the process of collision avoidance is proposed,and the behavior of target ship is divided into two types: assistant action and resistance action,and the trigger conditions,basic states and state transition conditions of the finite state machine are defined,which enable collision avoidance decisions to be improved from semi-dynamic scenarios to fully dynamic scenarios;Finally,based on the “one-step dynamic characteristics” of the Markov decision theory and the front-back end closed-loop design framework,a ship collision avoidance decision-making process considering the constraints of multiple navigation characteristic elements is proposed.Based on the electronic chart simulation environment,this method is tested in 10 typical ship collision avoidance scenarios.The experimental results prove that this method can effectively deal with multiple types of collision avoidance scenarios where two ships,multiple ships and target ships taking action,and verifies the rationality and advancement of this method.(3)From the perspective of “global”,this dissertation proposes the multi-vessel colaborative collision avoidance strategy in mixed scenarios.Considering that with the development of maritime autonomous ship,ships with different levels of autonomy will coexist and then form a multi-modal special mixed navigation scenario.In view of the characteristics of different decision-making agents in this scenario,such as different cognition,different decision-making logic,and interactive coupling between decision-making units,this dissertation uses cooperative game theory for the first time to solve the multi-ship colaborative collision avoidance problem in mixed scenarios.First of all,in order to express the relationship between ships more clearly,this dissertation constructs a global dual-matrix model(including the collision risk matrix and the encounter situation matrix),proposes a dual-matrix dynamic update and maintenance mechanism,and combined with the network topology theory,the relationship between ships in the mixed navigation scenarios is abstractly represented as a multi-ship cooperative network,so as to realize the segmentation of the game sub-alliance and determine the cooperative subject;Afterwards,according to the different dimensions of the sub-alliances,different cooperation revenue functions are respectively constructed.According to the principle that the contribution of each ship to the alliance revenue is proportional to the revenue distribution of each ship in the alliance,the Shapley-value method is used to calculate the collision avoidance responsibility ratio of each ship in this multi-ship cooperative game,based on this,the solution set of multi-ship cooperative game is obtained;Finally,based on the results of the division of collision avoidance responsibilities,the generalized reciprocal velocity obstacle model is used to solve the multi-ship cooperative collision avoidance stragety.This collaborative strategy can satisfy group rationality and individual rationality at the same time,and can promote the participation of ships in mixed scenarios and form a stable alliance.The simulation results verify the effectiveness of the proposed multi-ship collaborative collision avoidance strategy.(4)Based on the above research results,this dissertation innovatively designs and develops the ship autonomous collision avoidance decision-making test system,complets the autonomous collision avoidance real ship test,further verified the collision avoidance decision-making method proposed in this dissertation,and realizes the collision avoidance test by “simulation” upgrade to “real ship”.First of all,this dissertation designs the architecture of the test system,and designs three functional modules including route keeping,collision avoidance decision-making and route return for the requirements of autonomous collision avoidance real ship test;Afterwards,a real ship test plan for autonomous collision avoidance is designed,and a specific process for the real ship test is built;Finally,the test system is installed on the 300 TEU container ship “Zhifei”,the autonomous collision avoidance is completed in the open waters of Qingdao coastal intelligent ship test field.